Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.perceptron.inc/llms.txt

Use this file to discover all available pages before exploring further.

​
What model name should I use?

Use model="perceptron-mk1" for Perceptron’s flagship vision-language model — supports image and video inputs, reasoning, and structured annotations (point, box, polygon, clip). Set provider="perceptron". For lighter-weight image-only variants (single-image or multi-image understanding), drop to the 0.2 series: model="isaac-0.2-2b-preview" for the standard 2B reasoning model or model="isaac-0.2-1b" for the compact edge-deployable 1B. The legacy model="isaac-0.1" is still supported for existing integrations.

​
Do I have to self-host?

No. The hosted API is the fastest way to ship. Self-hosting is optional for the 0.x series — pull the isaac-0.2-2b-preview weights, the isaac-0.2-1b weights, or the isaac-0.1 weights if you need to deploy on-prem or at the edge. Perceptron Mk1 is hosted-only.

​
How do I stream partial results?

Use the stream=True flag (or the perceptron.question helper) to receive text.delta and points.delta events as the model reasons.

​
Why am I getting AuthError?

The SDK found no usable API key. Double-check that PERCEPTRON_API_KEY is set before running your script, or pass configure(api_key="sk-live-...") inline. If you use .env, call load_dotenv() before invoking configure. Still stuck? Contact support with logs or share a repro in our Discord.